Integrated multiblock data analysis for improved understanding of grape maturity and vineyard site contributions to wine composition and sensory domains

Understanding the contributions of site, cultivar and grape maturity to the final wine composition and sensory qualities has increased over the past decade as researchers attempt to define contributors to terroir. In this investigation we focused on determining the relative contributions of multiple factors related to vine growth, berry maturity and site mesoclimate to finished wine composition and sensory profiles for Shiraz and Cabernet Sauvignon for two vintages. Grape maturation was monitored using a berry sugar accumulation model and wines made from three harvests at specific grape maturation stages. Comprehensive targeted grape and wine analysis of amino acids, carotenoids, sugars, organic acids, anthocyanins and grape volatile compounds were combined with targeted wine volatile and non-volatile chemical measures of composition and results for sensory descriptive analysis. Multiple chemometric models consisting of balanced sample sets derived from the pool of Shiraz and Cabernet Sauvignon samples were used in an ANOVA multiblock framework with orthogonal projection to latent structures (Boccard and Rudaz, 2016) to elucidate the relative importance of the design factors in each model. In this approach multiple data matrices are derived from the experimental design factors. Once an effect matrix is determine this is subtracted from the original data matrix to obtain pure effects and interaction submatrices with structured orthogonal data that summarises the design main effects and interactions. A response matrix is derived from the positive eigenvalues associated with each effects matrix and residuals are then added to each submatrix prior to kernel OPLS decomposition. Model performance evaluated from residual structure ratio (RSR), goodness of fit (R2Y) and permutation testing identified the significant factors associated with each model. Projection of sample scores associated with significant factors against scores of the residual matrix was used to visually assess sample clusters with confidence intervals based on Hotelling T2 metrics. Loadings from significant experimental factors were extracted from each model and use for hierarchical cluster analysis (HCA) with Euclidean distance measures and Wards grouping criteria. Prior to HCA model scores and loadings were rotated to ensure consistent presentation of factor levels in scores plots for each model. A conservative interpretation of loadings heat maps was considered appropriate and a summary heat map for explanatory factors is presented that enable interpretation of the impact of cultivar, site, grape maturity and region on grape and wine composition. The integrated data driven approach used in this investigation may be of assistance for other investigators for omics based experiments